/* fprobe: Foundry OS's libc self-check (milestone M2).
 *
 * A small, deterministic cousin of Clarke's mprobe2 probe
 * (clarke-scratch/exp16..exp19/mprobe2.c): it checks what the C library
 * gets back from the Rustbelt kernel -- the VALUES, not just "returned 0".
 * Every check prints one line ending in "ok" or "FAIL"; the last line is
 *   fprobe: all N checks passed        (exit code 0), or
 *   fprobe: F of N checks FAILED       (exit code F).
 * Timing checks leave room for Rustbelt's ~55 ms timer tick. The
 * benchmark-style checks of mprobe2 (which can fail on a busy host) are
 * left out on purpose: `make test` must not be flaky.
 *
 * Built with toolchain/foundry-cc (static-PIE). Runs on Linux too:
 *   toolchain/foundry-cc -O2 -pthread c/fprobe.c -o out/fprobe && out/fprobe
 * (a few checks are Rustbelt-specific and accept Linux's answer as well).
 */
#define _GNU_SOURCE
#include <dirent.h>
#include <elf.h>
#include <errno.h>
#include <fcntl.h>
#include <ftw.h>
#include <poll.h>
#include <pthread.h>
#include <signal.h>
#include <stdint.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/auxv.h>
#include <sys/epoll.h>
#include <sys/mman.h>
#include <sys/stat.h>
#include <sys/timerfd.h>
#include <sys/utsname.h>
#include <sys/wait.h>
#include <time.h>
#include <unistd.h>

static int fails, checks, on_rustbelt;

/* One check: print its name, then ok or FAIL (with errno when it failed). */
static void chk(const char *what, int ok)
{
	int e = errno;
	checks++;
	if (!ok)
		fails++;
	printf("  %-46s %s", what, ok ? "ok" : "FAIL");
	if (!ok)
		printf(" (errno %d %s)", e, strerror(e));
	printf("\n");
	errno = 0;
}

static long long now_ms(void)
{
	struct timespec t;
	clock_gettime(CLOCK_MONOTONIC, &t);
	return t.tv_sec * 1000LL + t.tv_nsec / 1000000;
}

static void ms_sleep(long ms)
{
	struct timespec t = { ms / 1000, ms % 1000 * 1000000L };
	nanosleep(&t, NULL);
}

/* ---- the program itself: ELF, auxv ------------------------------------- */
extern const char __ehdr_start[];

static void elf_and_auxv(void)
{
	printf("fprobe: -- ELF and auxv\n");
	const Elf64_Ehdr *me = (const void *)__ehdr_start;
	const Elf64_Phdr *ph = (const void *)(__ehdr_start + me->e_phoff);
	int interp = 0;
	for (int i = 0; i < me->e_phnum; i++)
		interp |= ph[i].p_type == PT_INTERP;
	chk("static-PIE: ET_DYN without PT_INTERP", me->e_type == ET_DYN && !interp);
	chk("AT_BASE == 0 (no dynamic loader)", getauxval(AT_BASE) == 0);
	chk("AT_PHDR == our header + e_phoff",
	    getauxval(AT_PHDR) == (unsigned long)__ehdr_start + me->e_phoff);
	chk("AT_PAGESZ 4096, AT_RANDOM present",
	    getauxval(AT_PAGESZ) == 4096 && getauxval(AT_RANDOM) != 0);
	chk("vDSO mapped (AT_SYSINFO_EHDR)", getauxval(AT_SYSINFO_EHDR) != 0);
}

/* ---- files in /tmp ------------------------------------------------------ */
static int rm_cb(const char *p, const struct stat *s, int f, struct FTW *w)
{
	(void)s, (void)f, (void)w;
	return remove(p);
}

static int count_dir(const char *d)
{
	DIR *dd = opendir(d);
	struct dirent *e;
	int n = 0;
	if (!dd)
		return -1;
	while ((e = readdir(dd)))
		if (strcmp(e->d_name, ".") && strcmp(e->d_name, ".."))
			n++;
	closedir(dd);
	return n;
}

static void files(void)
{
	printf("fprobe: -- files (/tmp)\n");
	const char *D = "/tmp/fprobe", *F = "/tmp/fprobe/a.txt";
	struct stat st, st2;
	char buf[64] = { 0 };
	nftw(D, rm_cb, 8, FTW_DEPTH | FTW_PHYS); /* leftovers from an earlier run */
	umask(022);
	chk("mkdir", mkdir(D, 0755) == 0);
	int fd = open(F, O_CREAT | O_TRUNC | O_RDWR, 0666);
	chk("open O_CREAT 0666 -> mode 0644 (umask 022)",
	    fd >= 0 && fstat(fd, &st) == 0 && (st.st_mode & 0777) == 0644);
	chk("write + lseek + read back",
	    write(fd, "hello world\n", 12) == 12 && lseek(fd, 6, SEEK_SET) == 6 &&
		    read(fd, buf, sizeof buf - 1) == 6 && strcmp(buf, "world\n") == 0);
	chk("ftruncate -> st_size 5", ftruncate(fd, 5) == 0 && fstat(fd, &st) == 0 && st.st_size == 5);
	chk("mtime is a real date (after 2020)", st.st_mtime > 1600000000);
	close(fd);
	chk("rename", rename(F, "/tmp/fprobe/b.txt") == 0 && stat(F, &st) == -1 && errno == ENOENT);
	errno = 0;
	chk("link: same inode, st_nlink 2",
	    link("/tmp/fprobe/b.txt", "/tmp/fprobe/c.txt") == 0 && stat("/tmp/fprobe/b.txt", &st) == 0 &&
		    stat("/tmp/fprobe/c.txt", &st2) == 0 && st.st_ino == st2.st_ino && st.st_nlink == 2);
	memset(buf, 0, sizeof buf);
	chk("symlink + readlink",
	    symlink("b.txt", "/tmp/fprobe/l") == 0 && readlink("/tmp/fprobe/l", buf, sizeof buf) == 5 &&
		    strcmp(buf, "b.txt") == 0);
	chk("lstat sees the link, stat follows it",
	    lstat("/tmp/fprobe/l", &st) == 0 && S_ISLNK(st.st_mode) && stat("/tmp/fprobe/l", &st2) == 0 &&
		    S_ISREG(st2.st_mode) && st2.st_size == 5);
	for (int i = 0; i < 20; i++) {
		snprintf(buf, sizeof buf, "/tmp/fprobe/f%02d", i);
		close(open(buf, O_CREAT | O_WRONLY, 0644));
	}
	chk("readdir lists all 23 entries", count_dir(D) == 23);
	chk("nftw(FTW_DEPTH) removes the tree (rm -r)",
	    nftw(D, rm_cb, 8, FTW_DEPTH | FTW_PHYS) == 0 && stat(D, &st) == -1);
	errno = 0;
	chk("open of a missing file -> ENOENT", open("/tmp/no/such", O_RDONLY) == -1 && errno == ENOENT);
}

/* ---- processes, pipes, signals ------------------------------------------ */
static volatile sig_atomic_t got;
static void on_sig(int s) { got = s; }

static void processes(void)
{
	printf("fprobe: -- processes, pipes, signals\n");
	int p[2], st = 0;
	chk("pipe", pipe(p) == 0);
	pid_t c = fork();
	if (c == 0) {
		close(p[0]);
		dprintf(p[1], "child %d", (int)getpid());
		_exit(7);
	}
	close(p[1]);
	char buf[32] = { 0 };
	int n = read(p[0], buf, sizeof buf - 1);
	close(p[0]);
	chk("fork: child writes into the pipe", n > 6 && strncmp(buf, "child ", 6) == 0 && atoi(buf + 6) == c);
	chk("waitpid: exit status 7", waitpid(c, &st, 0) == c && WIFEXITED(st) && WEXITSTATUS(st) == 7);
	c = fork();
	if (c == 0) {
		execl("/no/such/program", "x", (char *)NULL);
		_exit(errno == ENOENT ? 3 : 4);
	}
	chk("execve of a missing file -> ENOENT", waitpid(c, &st, 0) == c && WEXITSTATUS(st) == 3);
	c = fork();
	if (c == 0) {
		pause();
		_exit(0);
	}
	kill(c, SIGTERM);
	chk("kill(SIGTERM) -> WIFSIGNALED",
	    waitpid(c, &st, 0) == c && WIFSIGNALED(st) && WTERMSIG(st) == SIGTERM);
	struct sigaction sa = { .sa_handler = on_sig };
	sigemptyset(&sa.sa_mask);
	sigaction(SIGUSR1, &sa, NULL);
	sigset_t s, old, pend;
	sigemptyset(&s);
	sigaddset(&s, SIGUSR1);
	sigprocmask(SIG_BLOCK, &s, &old);
	got = 0;
	raise(SIGUSR1);
	sigemptyset(&pend);
	chk("blocked signal stays pending", got == 0 && sigpending(&pend) == 0 && sigismember(&pend, SIGUSR1));
	sigprocmask(SIG_SETMASK, &old, NULL);
	chk("unblocking runs the handler", got == SIGUSR1);
}

/* ---- threads, memory ----------------------------------------------------- */
static __thread int tls = 7;
static long counter;
static pthread_mutex_t lock = PTHREAD_MUTEX_INITIALIZER;

static void *worker(void *arg)
{
	int ok = tls == 7;
	tls = 100 + (int)(long)arg;
	for (int i = 0; i < 20000; i++) {
		pthread_mutex_lock(&lock);
		counter++;
		pthread_mutex_unlock(&lock);
	}
	return (void *)(long)(ok && tls == 100 + (int)(long)arg);
}

static void threads_and_memory(void)
{
	printf("fprobe: -- threads and memory\n");
	pthread_t t[4];
	long all = 1;
	for (long i = 0; i < 4; i++)
		pthread_create(&t[i], NULL, worker, (void *)i);
	for (int i = 0; i < 4; i++) {
		void *r;
		pthread_join(t[i], &r);
		all &= (long)r;
	}
	chk("4 threads: own copy of __thread data", all && tls == 7);
	chk("4 threads x 20000 mutex increments = 80000", counter == 80000);
	char *m = mmap(NULL, 1 << 20, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
	int zero = m != MAP_FAILED;
	for (int i = 0; zero && i < 1 << 20; i += 4096)
		zero = m[i] == 0;
	chk("mmap 1 MiB anonymous, zero-filled", zero);
	if (m != MAP_FAILED) {
		memset(m, 0x5a, 1 << 20);
		char *g = mremap(m, 1 << 20, 4 << 20, MREMAP_MAYMOVE);
		chk("mremap grow 1 -> 4 MiB keeps the data",
		    g != MAP_FAILED && g[0] == 0x5a && g[(1 << 20) - 1] == 0x5a && g[(4 << 20) - 1] == 0);
		munmap(g != MAP_FAILED ? g : m, g != MAP_FAILED ? 4 << 20 : 1 << 20);
	}
	char *big = malloc(8 << 20), *small = malloc(100);
	chk("malloc 8 MiB + 100 bytes", big && small);
	if (big)
		big[(8 << 20) - 1] = 1;
	free(big);
	free(small);
	char out[64];
	snprintf(out, sizeof out, "%.3f %s %d", strtod("2.5e1", NULL) / 8, "x", -42);
	chk("printf %f / strtod", strcmp(out, "3.125 x -42") == 0);
}

/* ---- clocks and timerfd (Rustbelt 0.20.0) -------------------------------- */
static int arm(int fd, int flags, long ms, long every_ms)
{
	struct itimerspec its = { { every_ms / 1000, every_ms % 1000 * 1000000L },
				  { ms / 1000, ms % 1000 * 1000000L } };
	return timerfd_settime(fd, flags, &its, NULL);
}

static void clocks_and_timerfd(void)
{
	printf("fprobe: -- clocks and timerfd\n");
	struct timespec a, b, rt;
	int back = 0;
	clock_gettime(CLOCK_MONOTONIC, &a);
	for (int i = 0; i < 100000; i++) {
		clock_gettime(CLOCK_MONOTONIC, &b);
		back += b.tv_sec < a.tv_sec || (b.tv_sec == a.tv_sec && b.tv_nsec < a.tv_nsec);
		a = b;
	}
	chk("100000 MONOTONIC reads never go backwards", back == 0);
	clock_gettime(CLOCK_REALTIME, &rt);
	chk("REALTIME is 2026 or later", rt.tv_sec > 1767225600L);
	long long t0 = now_ms();
	ms_sleep(200);
	long long slept = now_ms() - t0;
	chk("nanosleep 200 ms takes 200-1000 ms", slept >= 195 && slept < 1000);

	/* One-shot timer, blocking read: waits ~150 ms, then the count is 1. */
	int fd = timerfd_create(CLOCK_MONOTONIC, TFD_CLOEXEC);
	chk("timerfd_create(MONOTONIC)", fd >= 0);
	uint64_t n = 0;
	t0 = now_ms();
	arm(fd, 0, 150, 0);
	int r = read(fd, &n, sizeof n);
	long long took = now_ms() - t0;
	chk("timerfd one-shot: read blocks ~150 ms, count 1", r == 8 && n == 1 && took >= 140 && took < 1000);
	struct itimerspec cur;
	chk("timerfd disarmed after firing",
	    timerfd_gettime(fd, &cur) == 0 && cur.it_value.tv_sec == 0 && cur.it_value.tv_nsec == 0);
	close(fd);

	/* Periodic timer: expirations add up while nobody reads. */
	fd = timerfd_create(CLOCK_MONOTONIC, TFD_NONBLOCK);
	errno = 0;
	chk("timerfd nonblocking + disarmed: EAGAIN", read(fd, &n, sizeof n) == -1 && errno == EAGAIN);
	arm(fd, 0, 60, 60);
	ms_sleep(700);
	n = 0;
	chk("timerfd 60 ms interval: 5-14 expirations in 700 ms",
	    read(fd, &n, sizeof n) == 8 && n >= 5 && n <= 14);
	printf("  (%llu expirations)\n", (unsigned long long)n);

	/* epoll wakes up when the timer fires. */
	int ep = epoll_create1(EPOLL_CLOEXEC);
	struct epoll_event ev = { .events = EPOLLIN, .data.u64 = 42 }, out;
	arm(fd, 0, 100, 0);
	read(fd, &n, sizeof n); /* drop anything left from the interval timer */
	epoll_ctl(ep, EPOLL_CTL_ADD, fd, &ev);
	t0 = now_ms();
	r = epoll_wait(ep, &out, 1, 3000);
	took = now_ms() - t0;
	chk("timerfd wakes epoll_wait", r == 1 && out.data.u64 == 42 && took < 1500);
	close(ep);

	/* Rustbelt doesn't do TFD_TIMER_CANCEL_ON_SET: it must say EINVAL. */
	struct itimerspec abs = { { 0, 0 }, { rt.tv_sec + 60, 0 } };
	errno = 0;
	r = timerfd_settime(fd, TFD_TIMER_ABSTIME | TFD_TIMER_CANCEL_ON_SET, &abs, NULL);
	chk(on_rustbelt ? "TFD_TIMER_CANCEL_ON_SET -> EINVAL (Rustbelt)" : "TFD_TIMER_CANCEL_ON_SET (Linux: works)",
	    on_rustbelt ? r == -1 && errno == EINVAL : r == 0);
	close(fd);
}

int main(void)
{
	struct utsname u;
	setvbuf(stdout, NULL, _IOLBF, 0); /* lines appear in order with the kernel's */
	on_rustbelt = uname(&u) == 0 && strcmp(u.sysname, "Rustbelt") == 0;
	printf("fprobe: Foundry OS libc self-check on %s %s\n", u.sysname, u.release);
	elf_and_auxv();
	files();
	processes();
	threads_and_memory();
	clocks_and_timerfd();
	if (fails)
		printf("fprobe: %d of %d checks FAILED\n", fails, checks);
	else
		printf("fprobe: all %d checks passed\n", checks);
	return fails;
}